The usage of recommender systems and their impact on everyday life has gained a lot of importance in recent years. The primary objective is to guide users to the discovery of new products and services by providing suggestions based on already known user interests and ratings. This is an essential feature in the digital world, since users tend to easily get overwhelmed by choice if given a large number of items to choose from.
There is a big variety of areas, in which recommender systems are used, like product recommendations for online shopping stores or artist respectively song recommenders for music streaming platforms.
Since the datasets for the respective tool can differ in size and structure, it is crucial to find the best suiting recommendation technique for rating predictions out of a big collection of different approaches and methods. A difficulty that can occur in recommender systems is the cold-start problem, which refers to an issue, in which it is challenging for the system to infer interactions between users and items due to insufficient information. This thesis introduces recommender systems in general and how they can be approached with different techniques as well as possible solutions for the cold start problem. The presented dataset describes a case of the cold-start problem, which needs to be resolved in recommender systems. For this purpose, the work is conducted to analyze whether the presented recommendation methods are suitable for the arXiv dataset, that will be the study object, or if it requires a different approach. The results achieved in this thesis can provide essential insights into how to deal with the cold start problem in datasets and which approaches are most suitable.